Help on Adaptive Quantizations

[Sirber]

Hi

I'd like to have more infos on how -aq works. Thanks!

[Haali]

It reduces qp in "flat" and ("bluish" or "dark") areas. Strength controls the amount of qp reduction, sensitivity controls how "flat" the area must be. Start with 0.5-0.7 strength an 18-20 sensitivity and increase strength up to 1.2 until the blocks are gone. If you use a bitrate target then flat areas are improved at the expense of other parts of the picture, if you use a constant qp, then flat areas simply get more bits. This is mostly useful if normal encode produces noticeable blocking artifacts.

[Sirber]

Thanks for the info!

I made a test with default (--aq) and I got more squares and artefacts than whitout.
source: anime 23 sec, 640x480@24fps, 368kbps.

I think I will not use --aq until I get better results. I'll post the files tomorrow.

[Haali]

At such low speeds it won't help. I think the reasonable range is 1-2mbps or a constant qp(20-26) encode.

[Sirber]

ok, thanks for the hints! :D

[IgorC]

when AQ is enable diference between quantizers of two frames isn´t entire number.
Will it be possible to set float number for Max. Q step?
Or it won't be convinient for speed/quality balance?

[akupenguin]

The fact that per-frame quality setting theoretically don't have to be integers has nothing to do with AQ. I've just never bothered to separate lambda from QP, since QP is a fine enough scale for most purposes.